A few newbie questions

If you're new to FAH and need help getting started or you have very basic questions, start here.

Moderators: Site Moderators, FAHC Science Team

Post Reply
acnash
Posts: 20
Joined: Sun Mar 11, 2018 4:34 pm
Hardware configuration: 32x Intel (R) Xeon (R) CPU E5-2687W 0 @ 3.10GHz
125.88GiB
1 Nvidia GeForce GTX 970
Mint Linux ver 18
Location: Oxford, UK
Contact:

A few newbie questions

Post by acnash »

Hi all,

I'm brand new to the Folding@Home community and I'm very excited to learn more. I have a few newbie questions concerning the underlying simulation calculations and I apologise in advance if they've been asked before. I'm a computational chemist with over 8 years of experience using Gromacs, so I was very pleased to see the all familiar "mdrun" executable in the FAHControl Log console. My questions are similarly related.

1) Gromacs kicks out .xtc and/or .trr files according to output coordinate and velocity settings in an .mdp file. Is FAHControl writing these files to my disk? If so, where, how often and I assume it must delete them at some stage.

2) After a task has finished running what is copied off my machine? A .cpt file?

3) Unless something changed in Gromacs 2018 that I don't know of, GPUs are used to handle the long-range electrostatic interactions, whilst interactions within the vdw cut-off are handled by the CPUs. If I have CPUs and a GPU on my folding box will only one task be assigned across both CPU and GPU or can separate tasks (two separate mdrun calculations) be allocated, one to the CPUs and another to the GPU i.e., two different project codes? I ask because my GPU is not great, and I know from running my own calculations that if I involve GPU calculations it will slow down the performance of my CPUs.

Thanks
Anthony
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: A few newbie questions

Post by bruce »

Yes, and No.

The FAHClient operates somewhat differently than a dedicated GROMACS installation but the processing is much as you describe. Each Donor can allocate one or more GPUs to process an assignment ("Work Unit" or WU) and the remaining CPUs to process a different WU. (Separate tasks process each assignment independently.) GROMACS is used to process assignments for the CPU and OpenMM is used to process assignments for a GPU (by way of OpenCL or perhaps CUDA).

GROMACS Work starts from downloaded .tpr and .xml files. Results consist of both a .trr file and a .xtc plus some log files, These are deleted after being uploaded. GROMACS processes the electrostatic forces using the CPUs and no GPU is required. [FAH originally started many years ago before Stream Computing was available for GPUs.]

A single trajectorY is broken up into manageable segments which are assigned serially to individuals.

All files are managed by the client. Completed work files are deleted as soon as the results are uploaded ... so your disk does not get full.

Your "not great" GPU won't slow down the CPU processing (and vice versa).
acnash
Posts: 20
Joined: Sun Mar 11, 2018 4:34 pm
Hardware configuration: 32x Intel (R) Xeon (R) CPU E5-2687W 0 @ 3.10GHz
125.88GiB
1 Nvidia GeForce GTX 970
Mint Linux ver 18
Location: Oxford, UK
Contact:

Re: A few newbie questions

Post by acnash »

Thanks, Bruce. Totally understood.
acnash
Posts: 20
Joined: Sun Mar 11, 2018 4:34 pm
Hardware configuration: 32x Intel (R) Xeon (R) CPU E5-2687W 0 @ 3.10GHz
125.88GiB
1 Nvidia GeForce GTX 970
Mint Linux ver 18
Location: Oxford, UK
Contact:

Re: A few newbie questions

Post by acnash »

Hi Bruce, a follow-up question concerning GPUs.

I'm fortunate to rely on top end nVidia cards at my institution, however, I also use my own machine (spec is in my profile) to perform simulations at home. They are usually side-projects and are left uninterrupted. Do you have any GPU suggestions or benchmarks where I would see an improvement in simulation performance for Folding@Home AND my own terminal exected mdrun jobs? I don't have the funds for a top of the range card, £300 is probably accessible. I am currently running a very cheap GeForce GT710 (192 cuda cores and 1 GB mem) as this is a patchwork machine and I needed the HDMI monitor connection; Folding@Home doesn't recognise the cards as eligible to use, I can understand why.

Thanks
Joe_H
Site Admin
Posts: 7856
Joined: Tue Apr 21, 2009 4:41 pm
Hardware configuration: Mac Pro 2.8 quad 12 GB smp4
MacBook Pro 2.9 i7 8 GB smp2
Location: W. MA

Re: A few newbie questions

Post by Joe_H »

Not certain how the current demand for cards by miners have affected sources you can purchase a GPU card from, but that amount puts you in the GTX 1050ti or GTX 1060 price range. Both are capable cards for folding.

As for the GT 710 you have, there are 3 different entries for one in the GPUs.txt file. However, due to a timing issue with the FAH installer on Linux it might not have detected a GPU and set the configuration for a GPU to false. Using FAHControl that value can be set to true and after a restart the client should download and install the latest GPUs.txt file. The same would be necessary before folding on a higher end card.

The other issue with Linux and GPU folding is the video drivers. You will need to install and use the nVidia proprietary drivers along with OpenCL support. With all that done even your 710 should fold, though it will be on the slower side.

To run alongside your own mdrun Gromacs jobs, you will probably need to reduce the number of CPU threads used. The way nVidia has implemented OpenCL in their drivers requires the use of a CPU core to handle setting up the data and OpenCL commands to transfer to and back from the GPU. The FAH experience with the folding cores based on Gromacs has been that the number of CPU threads needs to be a multiple of 2, 3 and sometimes 5. Multiples of primes greater than or equal to 7 have had problems.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: A few newbie questions

Post by JimboPalmer »

I get about 175k PPD with a low profile GTX 1050ti. I think a full height card would do better. The GTX 1050ti does not need additional power connectors.

They look to be under £175 before VAT and shipping https://www.newegg.com/global/uk/Produc ... 6814137055 as an example.

Scan also had models under £175 https://www.scan.co.uk/products/asus-ge ... 2mhz-boost

More powerful folders will need extra connectors, so be sure what your PSU offers.

(The GT 710 is a rebadged GT 520)
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
Joe_H
Site Admin
Posts: 7856
Joined: Tue Apr 21, 2009 4:41 pm
Hardware configuration: Mac Pro 2.8 quad 12 GB smp4
MacBook Pro 2.9 i7 8 GB smp2
Location: W. MA

Re: A few newbie questions

Post by Joe_H »

JimboPalmer wrote:(The GT 710 is a rebadged GT 520)
You are probably thinking of the GT 610 or GT 705 which are rebadged GT 520's. The GT 710 is based on Kepler, not Fermi like the GT 520/610/705.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
acnash
Posts: 20
Joined: Sun Mar 11, 2018 4:34 pm
Hardware configuration: 32x Intel (R) Xeon (R) CPU E5-2687W 0 @ 3.10GHz
125.88GiB
1 Nvidia GeForce GTX 970
Mint Linux ver 18
Location: Oxford, UK
Contact:

Re: A few newbie questions

Post by acnash »

Thanks very much all, extremely useful information. The last time I had shown so much interest in graphics cards was when Doom 3 and Halflife 2 were released! Long ago. I'll try and get my 710 folding whilst I source a new card. Thanks again.
acnash
Posts: 20
Joined: Sun Mar 11, 2018 4:34 pm
Hardware configuration: 32x Intel (R) Xeon (R) CPU E5-2687W 0 @ 3.10GHz
125.88GiB
1 Nvidia GeForce GTX 970
Mint Linux ver 18
Location: Oxford, UK
Contact:

Re: A few newbie questions

Post by acnash »

Just for interest, I've got the cheap GeForce GT 710 LP running (this was initially just for a HDMI socket). I'm not sure how best to report folding efficacy, however, the Estimated PPD is 7080, whereas the neighbour 31 Intel Xeon 3.10GHz CPUs have an Estimated PPD of 178,731 but can fluctuate to as much as 200,000.
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: A few newbie questions

Post by bruce »

If you need a LP replacement, I am very happy with my GTX 1050Ti LP. It seems to be the highest performance LP GPU and it's not high enough on the coin miners list to be constantly out-of-stock. The non-Ti version is a close second.
Post Reply